The relationship between dynamic and pseudo-thermodynamic measures of the

glass transition temperature in nanostructured materials

Jayachandra Hari Mangalara, Mark E. Mackura, Michael D. Marvin, and David S. Simmons

Citation: The Journal of Chemical Physics 146, 203316 (2017); doi: 10.1063/1.4977520
View online: http://dx.doi.org/10.1063/1.4977520
View Table of Contents: http://aip.scitation.org/toc/jcp/146/20
Published by the American Institute of Physics
 THE JOURNAL OF CHEMICAL PHYSICS 146, 203316 (2017)

The relationship between dynamic and pseudo-thermodynamic measures

of the glass transition temperature in nanostructured materials
Jayachandra Hari Mangalara, Mark E. Mackura, Michael D. Marvin, and David S. Simmonsa)
Department of Polymer Engineering, The University of Akron, 250 South Forge St, Akron,
Ohio, 44325-0301 USA
(Received 10 October 2016; accepted 13 February 2017; published online 10 March 2017)

Despite decades of research on the effects of nanoconfinement on the glass transition temperature T g ,
apparent discrepancies between pseudothermodynamic and dynamic measurements of these effects
have raised questions regarding the presence of long-ranged interfacial dynamic gradients in glass-
forming liquids. Here we show that these differences can be accounted for based on disparities
in these methods weightings over local T g s within an interfacial gradient. This finding suggests
that a majority of experimental data are consistent with a broad interfacial dynamic interphase in
glass-forming liquids. Published by AIP Publishing. [http://dx.doi.org/10.1063/1.4977520]

INTRODUCTION dynamics under nanoconfinement due to free volume diffusion

effects.47
Beginning in the 1990s, numerous studies have reported
Recently, we and others have suggested that other appar-
alterations in the glass transition temperature T g of glass-
ent discrepancies in the behavior of nanoconfined materials
forming liquids under nanoconfinement,111 as measured via
as probed by distinct metrologies and predicted by computa-
ellipsometry and calorimetry. These effects are observed
tion can be resolved by considering differences in the ways
in polymer thin films, small-molecule liquids confined to
in which these methods probe or average over near-interface
pores,4,6 colloids,12 multi-nanolayered films,13,14 and poly-
gradients in dynamics.33,44,5358 Examples include differences
mer nanocomposites.1519 They are accompanied by measure-
between ellipsometry and Debye-Waller factor measurements
ments of alterations in other properties including modulus2022
from neutron scattering,53 differences in the apparent range of
and chain diffusion rates23 under nanoconfinement. Recent
the dynamic interface as determined via different experimental
evidence11 suggests that the same effects play an important
and computational methods,33,54 and differences between flo-
role in the behavior of ionomers,24,25 semi-crystalline poly-
rescence and ellipsometric measurements.56,57 Most recently,
mers, and block copolymers,2628 making them ubiquitous
the elastically collective nonlinear Langevin equation theory
contributors to the behavior of modern materials. These effects
for glass formation has been shown to lead to distinct pre-
have most commonly been attributed to the presence of long-
dictions for the values of dynamic and thermodynamic T g s
ranged gradients in dynamics and T g reported in the vicinity
in freestanding thin films,59,60 consistent with the observation
of interfaces in these materials.9,17,2937 This would appear to
of Lipson and Milner that mean film T g s may not reflect a
represent a finding of extreme scientific and technological rele-
democratic sampling of local dynamics.61 Here we consider
vance: it has been suggested that these shifts provide a window
the question of whether a difference in weightings over an
into underlying scales of cooperative dynamics in supercooled
interfacial T g gradient can account for apparent discrepan-
liquids,2,29,30,3336,3844 and they are implicated in determin-
cies between dynamic and pseudo-thermodynamic measures
ing material properties in applications from membranes to
of T g in nanoconfined materials in a general manner, with-
microelectronics.
out the assumption of any particular model for film dynamics
However, this conclusion has been complicated by work
and without resort to the proposed experimental anomalies or
reporting that T g , when measured by dynamic methods such
novel mechanisms described above.
as dielectric spectroscopy,45 frequently exhibits much weaker
shifts under nanoconfinement than is observed via pseudo-
thermodynamic measurements.4649 This apparent discrep- MOLECULAR DYNAMICS SIMULATIONS
ancy has raised questions of whether nanoconfinement and
interfaces indeed impose long-ranged alterations in the dynam- We begin by verifying that these apparently discrepant
ics of glass-forming liquids. Attempts to explain it have findings can be observed in a single system, in the absence of
included suggestions of inadequate film annealing,50 sol- confounding factors such as solvent contamination. To do so,
vent contamination,51 differences in effective time scale,51,52 we perform molecular dynamics simulations of glass forma-
flaws in data analysis, and the proposition that the pseudo- tion in bulk and freestanding-film geometries of a linear, unen-
thermodynamic T g is entirely decoupled from underlying tangled bead-spring polymer based on the model of Kremer
and Grest,62 which has been widely employed to study poly-
mer glass formation.8,16,32,6365 Because the fragility (thermal
a)
Author to whom correspondence should be addressed. Electronic mail: abruptness) of the glass transition has been implicated in mod-
dsimmon@uakron.edu ulating the nature and strength of nanoconfinement effects,42,66

0021-9606/2017/146(20)/203316/8/$30.00 146, 203316-1 Published by AIP Publishing.

203316-2 Mangalara et al.
FIG. 1. Extrapolated bulk state fragility index mE , computed based on a
Vogel-Fulcher-Tammann extrapolation to the experimental time scale of the
glass transition, vs bond length. Error bars correspond to standard deviations
over four independent runs.
we also include several variations of this model probing a
substantial fragility range in order to improve the robustness
and generalizability of our results. Specifically, by shortening
the length of backbone bonds while leaving all other param-
eters of the model the same,67 we improve the segmental
packing of the model and thereby reduce bulk-state fragility
substantially (see Figure 1 and the supplementary material).
While this model system does not correspond to any particu-
lar polymer, this approach allows us to ensure that results are
representative of the broad range of fragilities of glass forma-
tion observed in experimental polymers. All simulated films
are approximately 13 nm thick in the vicinity of T g , based on
a common mapping of order one Lennard-Jones distance unit
to 1 nm.8,27,62,68
SIMULATION PROTOCOL
We perform molecular dynamics simulations of model
bead-spring polymers in the bulk state and in a free-
standing film geometry. Simulations employ attractive bead-
spring model polymers extended from the earlier model
of Kremer and Grest,62,69 which has been extensively
employed to study the polymer glass formation behav-
ior.17,7074 Prior work indicates that this model exhibits signif-
icant alterations in the dynamics and glass formation behavior
under nanoconfinement comparable to those in experi-
ment.8,14,32,33,42,43,63,75,76 This model offers the advantage of
a relatively small onset-time scale of these effects as com-
pared to some experimental systems52,77,78 while retaining
qualitative features of experimental nanoconfinement effects,
enabling the study of nanoconfinement-induced alterations
in dynamics and T g within computationally accessible time
scales.
Within this model, non-bonded interactions are given by
a 12-6 Lennard-Jones (LJ) potential,
ij  12  ij  6
" #
Eij = 4 ij , (1)
r r
where ij and ij are Lennard-Jones energy and range param-
eters, respectively, and where the interaction is truncated and
shifted to 0 at r = 2.5.79 Bonded beads interact via the Finitely
Extensible Nonlinear Elastic (FENE) potential,
J. Chem. Phys. 146, 203316 (2017)
! 2
r
EFENE = 0.5KFENE R02 ln 1
R0
"   12   6 #
+ 4 + , (2)
r r
where the first term sets a maximum bond length and the sec-
ond term describes the Lennard-Jones repulsion. This bonding
model has the advantage that it largely prevents aphysical
chain-crossing events.
The polymer in its bulk state consists of a melt of linear
20-bead chains, well below the entanglement limit of 85 for
this model.80,81 Non-bonded interactions between monomers
employ standard values of pp and pp equal to 1 and mass
m = 1 (where the subscript p denotes the polymer repeat
unit). In order to simulate glass formation in polymers span-
ning a range of bulk fragility of glass formation, we employ a
series of models varying in terms of the length of the backbone
FENE bonds between beads. To do so, in addition to a standard
FENE polymer with K = 30 and R0 = 1.5, we simulate poly-
mers in which these FENE bond parameters are systematically
altered (Table I) to yield the bulk fragility values as shown in
Figure 1.
Simulations are carried out using the 22 February 2013
release of LAMMPS82 (Large-scale Atomistic/Molecular
Massively Parallel Simulator) for all simulated systems except
the standard FENE polymer, for which we used the 5 Septem-
ber 2014 release of LAMMPS. All quantities are in dimension-
less LJ units. Simulations employ the Nose-Hoover thermostat
and barostat, as implemented in LAMMPS, with dampen-
ing coefficients for both thermostat and barostat equal to 2
LJ ,73 where LJ is the LJ unit of time. All simulations in
the NPT ensemble are performed at pressure P = 0. We
employ the rESPA (reversible reference system propagator
algorithm) time integrator,83 allowing a different time step to
be used for the integration of bonded and non-bonded inter-
actions. A 4:1 ratio is employed between non-bonded and
bonded time steps, with an outer (non-bonded) time step of
0.01 LJ .
In order to quantify glass formation behavior, we prepare
simulated systems by quenching the material from high to
low temperature and then performing a post-quench anneal at
each temperature to obtain a well-defined temperature range
of in-equilibrium configurations.33,73 Specifically, random ini-
tial configurations are generated using PACKMOL,84 with the
polymer chains filling the entire box for the bulk simulation or
placed into a freestanding film configuration for the film geom-
etry. The bulk system is then equilibrated at a temperature of
T = 1.5 in the NPT ensemble for 104 LJ . By comparison, the
T g s of all model systems considered in this work are below
T = 0.45, such that all systems are initially equilibrated at a
TABLE I. FENE parameters for the changing bond length.
Bond length FENE parameters
0.96 R0 = 1.5; = 1.0
0.88 R0 = 1.4; = 0.9
0.83 R0 = 1.35; = 0.85
0.79 R0 = 1.3; = 0.8
203316-3 Mangalara et al. J. Chem. Phys. 146, 203316 (2017)

TABLE II. NPT equilibration time after quench. with the dominance of short-range interactions in the glass
transition. Also consistent with recent work,14,33,76 the
Temperature range Equilibration length
structural relaxation time is defined as the time at which
1.0 T 1.5 10 000 LJ F s (k,t) decays to 0.2, employing a fit to the Kohlrausch-
0.7 T < 1.0 20 000 LJ Williams-Watts (KWW) stretched exponential,87,88
0.1 T < 0.7 100 000 LJ   
Fs (k, t) = A exp t/ (4)

to F s (k,t) at times after the inertial fast relaxation process73

temperature greater than 3T g . The film system is subject to the
(1 LJ ) is complete, for data smoothing and interpolation.
same initial equilibration temperature and period, albeit in the
A fit of the relaxation time versus temperature to the Vogel-
NVT ensemble, allowing the system to settle into a film con-
Fulcher-Tammann8991 (VFT) equation,
figuration with large void spaces above and below the film, !
such that an effective NPT ensemble is probed despite the DT0
= 0 exp , (5)
nominal use of a constant volume boundary condition. Peri- T T0
odic boundary conditions are employed in all cases. These
is then employed for data smoothing prior to the determination
initial configurations are quenched in the respective ensem-
of characteristic temperatures of glass formation as described
bles for bulk (NPT) and thin film (NVT) systems at a rate of
in the section titled Simulation results.
105 T /LJ to a minimum temperature of T = 0.1 while saving
Approximate experimental-time scale fragilities of glass
configurations at regular intervals. Configurations at each tem-
formation (shown, for example, in Figure 1) are then deter-
perature are then subjected to an additional equilibration in the
mined for these models based on the value of the kinetic
NPT ensemble for the bulk and the NVT ensemble for thin film
fragility index,
systems for a duration specified in Table II. Each simulation for
bulk polymer systems is then slightly rescaled to the average
log 

density at that temperature as determined from an average over m=  .   , (6)
the second half of the equilibration runs for four independent Tg T 
simulations and subject to an additional equilibration for a
period of 103 LJ in the NVT ensemble prior to the collection
of data in the NVT ensemble. For thin films systems, we
directly proceed to the data collection stage in the NVT
ensemble.
This annealing schedule is chosen to ensure an equili-
bration duration of at least 100 , where is the segmental
relaxation time at a specified temperature, up to a maxi-
mum computationally tractable equilibration time of 105 LJ .
Throughout this paper, relaxation-time data are based purely
on temperatures satisfying the above 100 equilibration-
length criterion. For the thin films studied here, this leads to
lowest in-equilibrium temperatures in the vicinity of 0.42 to
0.43.
ANALYSIS PROTOCOL
Structural relaxation dynamics are quantified based on
the self-part of the intermediate scattering function85 which is
given by
N
  1 XD f   gE
Fs ~k, t = exp i~k ~rj (t) ~rj (0) , (3)
N j
where ~rj (t) is the position of the jth particle at time t and
N is the number of particles. Relaxation times are based on
an average over wave-vectors corresponding to a wavenum-
ber of k = 7.07, chosen to be comparable to the first peak
in the monomer structure factor. We note that there has
been little work systematically studying the question of how
nanoconfinement effects vary as a function of k, but a choice
comparable to the scale of the segmental structure is consis-
tent with extensive prior work on the glass transition under
nanoconfinement14,24,25,32,33,42,43,76,86 and is also consistent
T =Tg
predicted by the VFT fit to each system at a time scale of 1014
LJ , approximately equivalent to the 100 s conventional time
scale of experimental glass formation.
In order to quantify local dynamics within the film, we
sort the particles into bins of 0.875 from the interface to
the center. We define the interface as the point at which the
density is equal to 0.5. We determine the precise location
of the interface by fitting the two halves of the box to two
sigmoidal functions and identifying the points where these
fit functions equal 0.5, approximately half the bulk density
at low temperature. We denote the position of the interface
by z = 0 and average all z-dependent data over the two
interfaces.
SIMULATION RESULTS
In each system, we compute T g via several dynamic and
pseudo-thermodynamic approaches, illustrated in Figure 2,
that have been employed in simulation studies76,92 to reflect
experimental T g conventions. A central perennial challenge
in the study of T g by simulation is the large difference in the
time scales accessible to molecular simulation as compared to
those conventionally associated with T g in the experiment.
In an effort to control for this gap to the extent possible,
we first employ two different dynamic conventions for T g .
A computational time scale dynamic T g , T g C , is defined
as the temperature at which goes to 103 LJ (the LJ
time unit), the longest relaxation time at which we consider
our simulation to be firmly in equilibrium. An extrapolated
experimental time scale dynamic T g , T g E , is defined as
the temperature at which these data extrapolate to 1014 LJ
(approximately 100 s), based upon a fit of in-equilibrium data
to the Vogel-Fulcher-Tammann relation.89,90 In general, we
203316-4 Mangalara et al.
FIG. 2. Determination of T g for the standard attractive Kremer-Grest model,
in the bulk (blue) and thin films (orange), via dynamic (top), dilatometric
(middle), and calorimetric (bottom) conventions.
view the qualitative agreement of these two conventions as
indicating a high likelihood that trends in T g at computational
time scales are well-representative of expected trends in T g
at experimental time scales, although this is by no means
a rigorous assurance. Finally, dilatometric and calorimetric
pseudo-thermodynamic T g s are defined as the temperatures
at which linear fits to the liquid and glassy regimes intersect
for the system volume and energy, respectively.
As shown in Figure 3(a), thin film pseudo-thermodynamic
T g s consistently exhibit a greater suppression relative to the
bulk than do dynamic T g s; the mean dynamic T g of the
four systems is suppressed by 5% while the mean ellipso-
metric T g is suppressed by 15% and calorimetric by 14%.
We also note that this difference between dynamic and pseu-
dothermodynamic methods is insensitive to the time scale at
which we consider the dynamics to fall out of equilibrium;
as shown by Figure 3(b), alternately defining the computa-
tional dynamic T g as the temperature at which = 104 , while
including data up to this time scale, or as the temperature at
which = 102 , while including only data up to this shorter
time scale, does not appreciably change the results. Experi-
mental anomalies, such as solvent contamination and residual
stresses from film spinning, are evidently not needed to observe
large differences between nanoconfinement-induced shifts in
dynamic and pseudo-thermodynamic T g s.
J. Chem. Phys. 146, 203316 (2017)
FIG. 3. Top: T g suppression relative to the bulk, based on extrapolated
dynamic (orange circles), computational dynamic (blue squares), dilatometric
(green diamonds), and calorimetric (red triangles) conventions, as a function
of bulk fragility computed as in Figure 1, with error bars representing the
standard deviation over four independent runs in corresponding colors. Bot-
tom: The same figure, but with computational dynamic T g computed in two
alternate ways to confirm the time scale-independence of these results. Open
diamonds indicate a computational dynamic T g defined at = 102 , including
only data with relaxation times less than or equal to this value. Open squares
indicated a computational dynamic T g defined at = 104 , including data
with relaxation times up to this value rather than the standard cutoff of 103 .
DISCUSSION
Dependence of T g gradient weighting on metrology
We now turn to the question of whether this differ-
ence between dynamic and pseudo-thermodynamic glass
transition temperatures can be explained by a difference in the
intrinsic sensitivity of dynamic and pseudo-thermodynamic
measures of T g to different parts of a local gradient in dynam-
ics near the interface. Rather than considering experimental
details of specific measurements, we focus on the underly-
ing difference between pseudo-thermodynamic and dynamic
measures of film T g : the former report the mean value of a
pseudo-thermodynamic property, while the latter report a mean
relaxation time. We compare the averages reported by these
classes of measurement to a linear arithmetic mean over the
local values of T g , given by
D E (z) Tg (z) dz
Tg =
, (7)
(z) dz
where (z) is the mean density in a plane parallel to the film
at a position (z), T g (z) is the local average T g within this
203316-5 Mangalara et al. J. Chem. Phys. 146, 203316 (2017)

plane, and brackets denote a film average. An assumption
of this mean has underpinned decades of nanoconfined T g
interpretation.2,44
Consistent with Figure 2, pseudo-thermodynamic deter-
minations of T g commonly employ a two-line construction,
wherein the temperature-dependence of a film-average ther-
modynamic property X is fit to separate linear functional forms
well above and well below T g ,
hXi i = hAi i T + hBi i ,
where i becomes l or g to denote properties in the liquid or glass
state, respectively, and Ai and Bi are parameters quantifying
the slope and intercept of the corresponding linear fits. The
measured pseudo-thermodynamic glass transition temperature
hTg ipt is then given by the equation
D E  D E . D E  question, we consider how dynamic measurements average
Tg = hBl i Bg Ag hAl i . (9)
pt
How should this be expected to average over an interfacial
gradient in T g ? For simplicity, we focus on an average over
a single interface at position z = 0 between an empty half-
space in the negative z and the material in the positive z. If we
consider any temperature well above or below the mean T g , the
film average value of X (density in dilatometry, or extensive
energy in calorimetry) is given by
T g is suppressed, than in the film center or near an attrac-
tive substrate (for example, see Figure 4), such that pseudo-
thermodynamic methods will generally yield a T g that is
weighted towards lower-T g regions of the film. This is an
intuitively reasonable finding. Experimentally, the change in
the slope of a pseudothermodynamic variable through T g is
often referred to as the strength of the glass transition.
In essence, Equation (14) therefore leads to the intuitive
conclusion that the mean film T g reflects a weighting by
(8)
the local pseudo-thermodynamic strength of T g through the
film.
Should we then conclude that differences between
dynamic and pseudo-thermodynamic measurements of T g in
nanoconfined systems emerge purely from the weighting of
pseudo-thermodynamic T g measurements? To answer this
over the film. These measurements probe a mean relaxation
time h i, which is nonlinearly related to T g ,93 and will
therefore yield a differently weighted average,
(z) (z) dz
h i =
. (15)
(z) dz

b To relate this to a mean inferred T g , we require a form for

1 the relationship between and T. For a sufficiently narrow
hXi i = lim Xi (z) dz. (10)
b b window around the glass transition, the temperature depen-
0 dence of can be reasonably well described by an effective
By definition, the relationship provided by Equation (8) Arrhenius temperature dependence,
applies both to film average or local in-film properties. Making
(z) = 0 (z) exp Ea (z) (kT ) ,
  
use of this fact gives (16)

b b where E a is an effective activation energy given by

1 1
hAi i = lim Ai (z) dz and hBi i = lim Bi (z) dz. (11) d ln
b b b b Ea = k 
. (17)
0 0 d 1 T
Combining Equations (9) and (11) yields Applying this equation at T g gives
b
g

Bg (z) Bl (z) dz Ea = kTg ln , (18)
0 0
D E
Tg = lim . (12)
pt b b  
Al (z) Ag (z) dz
0
Applying Equation (9) to local dynamics in the film gives

D E b
Tg = lim Tg (z) wpt (z) dz, (13)
pt b
0
where wpt is a weighting function with the properties of a
probability distribution function, given by
 
Al (z) Ag (z)
wpt = . (14)
b 
Al (z) Ag (z) dz
0
Equation (14) is the key result of the above development:
the mean film T g determined via any pseudo-thermodynamic
method is a film average of local T g s, weighted by the local
difference between the liquid and glass-state temperature
FIG. 4. Density as a function of temperature for the standard Kremer-Grest
derivatives of the corresponding thermodynamic variable. thin film polymer for layers of particles at mean distances from the interface
This difference is often larger near a free surface, where indicated in the legend in LJ distance units.
203316-6 Mangalara et al. J. Chem. Phys. 146, 203316 (2017)

where g is the fixed relaxation time conventionally defining dynamic T g s will tend to be higher than corresponding
the location of the dynamic T g (commonly 100 s) and 0 is pseudo-thermodynamic T g s under nanoconfinement. Given
an adjustable parameter not to be equated with the genuine the large fraction of studies focusing on systems exhibiting
high-temperature relaxation time. Taking the above equations suppressed T g under nanoconfinement, it is to be expected
to hold locally at any plane in the film with position depen- that an apparently weaker T g -nanoconfinement effect would
dent parameters T g (z) and 0 (z) and substituting back into commonly be observed by dynamic methods. We note that this
Equation (15) then yields explanation does not require any difference in the underlying
gradient of dynamics probed by these methods.
g
 
Tg (z)
(z) 0 (z) exp T ln 0 (z) dz
h i =

, (19) Test of T g weighting in simulation

(z) dz If this reasoning is correct, then performing the averages

given by Equations (13) and (22) over a local dynamic T g
which provides an equation for the mean film relaxation time gradient calculated for these simulations should yield inferred
in terms of local T g s. In order to obtain an equation for the mean film T g s in closer agreement with the pseudothermo-
mean film T g as determined via a dynamic measurement, we dynamic and dynamic overall film T g s. To perform this test,
consider the application of Equation (18) to the film average we first determine the segmental relaxation time on a layer-
relaxation behavior and rearrange to give by-layer basis as described in the Molecular Dynamics Simu-
D E ,
g
! lations section. We then determine T g c for each layer via the
Tg = hEa i k ln . (20) same approach employed for the overall film. The resulting
d h0 i
T g gradients are shown in Figure 5 top. As in prior sim-
Because the Arrhenius relations employed to obtain ulation studies, these systems exhibit a smooth suppression
these equations are defined at T g rather than at high tem- in T g upon approaching the free surface. We note that since
perature, local Arrhenius parameters can be related to the this gradient is calculated dynamically from bins of discrete
local temperature-dependent fragility of glass formation m(z), thickness, it already incorporates a portion of the dynamic
defined as weighting. Use of Equations (13) and (22) will therefore still
d log (z) 1 Ea (z) 1 g tend to yield results somewhat biased towards the (higher)
m (z) =  .  = = ln .
d Tg (z) T Tg (z) ln 10 k ln 10 0 (z) mean dynamic T g . An exceptionally high spatial resolution
gradient would eliminate this issue but is prohibited by poor
(21) statistical sampling in very small bins.
Making use of the last of these equalities, while employing As shown by Figure 5 bottom, a simple density-weighted
Equation (17) at the level of the mean-film relaxation time to numerical average over this gradient (Equation (7)) yields
determine hEa i, and with Equation (19) providing the film- a mean film T g that is intermediate between the mean film
mean relaxation time, then leads to dynamic and pseudothermodynamic values, consistent with
 the weighting arguments above. We then can compare this to
D E m (z) an average over this gradient employing the dynamic weight-
Tg = Tg (z) f (z) , (22) ing given by Equations (22) and (23). We note that because
d hmi

this equation involves the mean film T g on both the left and
where right sides, it requires an iterative numerical solution. We
 D E   
T (z)
  also find that the density weighting in Equation (23) is neg-
z, T = Tg exp m (z) Tg 1 ln 10 dz ligible compared to the other factors for these freestanding
d h g id
f (z) =      , films and therefore neglect it. As shown by Figure 5 bot-
D E  Tg (z)
z, T = Tg d exp m (z) hTg i 1 ln 10 dz tom, the resulting inferred dynamic film T g is considerably
d
higher than the simple linear arithmetic average (increasing
(23)
to a 3% reduction relative to bulk) and similar to the over-
and where we have neglected the temperature-derivative of all film dynamic glass transition temperature, consistent with
the local density for simplicity. Here f (z) is again a weighting the above arguments. On the other hand, applying the pseu-
function with the properties of a probability distribution func- dothermodynamic weighting given by Equation (13) shifts
tion. The local fragility m(z) is defined as per Equations (18) the gradient-average T g towards the mean-film pseudother-
and (20) at the local value of T g (z), and hmi is the mean film modynamic T g , decreasing to a 10% reduction relative to
fragility defined at the mean film T g , which averages over the the bulk. As discussed above, the overshoot relative to the
dynamic gradient in a manner we have previously shown.76 mean dynamic T g measurement and undershoot relative to
Equations (22) and (23) indicate that the mean film T g the mean film pseudothermodynamic T g result from the fact
determined via dynamics is weighted towards regions of the that these discrete gradients already contain some dynamic
film with higher T g and higher fragility. weighting.
These derivations indicate that mean dynamic measures This analysis indicates that simply considering these dif-
of T g are weighted towards the slower-relaxing parts of ferent weightings can account for very large differences in
nanostructured materials, whereas mean pseudothermody- T g shifts under nanoconfinement as probed by dynamics vs
namic measurements of T g are weighted towards the faster- pseudo-thermodynamics. This reinforces the proposition that
relaxing parts of these materials. For this reason, mean the difference between the two measurement classes is a matter
203316-7 Mangalara et al.
FIG. 5. Top: Gradients in computational dynamic T g for systems with bond
lengths 0.96 (blue), 0.88 (orange), 0.83 (green), and 0.79 (purple). Bottom:
T g suppression vs. bulk fragility, with mean film T g computed via several
methods. Solid blue squares and solid green diamonds correspond to the mean
film computational dynamic and dilatometric T g s as in Figure 3, respectively.
Solid black circles employ mean film T g s computed via a linear arithmetic
average (Equation (7)) over the T g gradients shown in the top panel. Open
green diamonds employ mean film T g s computed by applying the pseudo-
thermodynamic weighting factor in Equations (13) and (14) to average over
the gradients in the top panel. Open blue squares employ T g s computed using
the dynamic weighting factor provided by Equations (22) and (23) to average
over the gradients in the top panel.
of weighting over a common gradient rather than a difference
in the underlying physics probed.
While this simulation test was performed in freestanding
films, one can expect the same weighting considerations to
apply in supported films as well. The surface gradient in such
films is widely understood to be comparable to surfaces on
freestanding films. On the other hand, if the substrate is suffi-
ciently rigid and attractive, a gradient towards slower dynamics
will then emerge at the substrate, as has been seen in prior
experimental and simulation work.14,29,32 The above weight-
ing equations indicate that such a regime will be overrepre-
sented in the dynamic average; an expected corresponding
reduction in liquid state thermal susceptibilities will lead to
an underrepresentation of this region in thermodynamic mea-
sures of T g . In essence, the quantity T g /T g ,bulk will remain
larger when measured by dynamic than by thermodynamic
methods. Thus, in scenarios in which the mean film T g drops
under nanoconfinement in supported films (as in a large frac-
tion of experimental studies of nanoconfinement), the effect
will appear to be larger when measured by thermodynamic
rather than dynamic methods.
J. Chem. Phys. 146, 203316 (2017)
CONCLUSIONS
In summary, while we cannot exclude a role for experi-
mental artifacts in specific measurement studies, we show that
these issues are not necessary to account for a general finding
of much weaker suppressions of T g under nanoconfinement
as probed by dynamic than pseudo-thermodynamic methods.
Moreover, measurements via both methods are intimately con-
nected to the underlying film dynamics: each reflects a differ-
ently weighted average over local gradients in these dynamics,
with pseudothermodynamic methods weighted towards low-
T g regions and dynamic methods weighted towards high-T g
regions. There is no way either weighting can yield a non-bulk-
like T g in the absence of an underlying gradient in T g , since
all other quantities involved in the average enter via weight-
ing functions that integrate to one. This finding suggests that
the majority of extant data are consistent with the presence of
long-ranged interfacial gradients in the segmental dynamics
and glass formation behavior.
Neither instrument details nor any model of the form of
interfacial gradients (such as the commonly employed two-
layer model) is necessary to arrive at these conclusions; rather
they emerge simply from the fact that dynamic measure-
ments fundamentally report a mean relaxation time while
pseudo-thermodynamic measurements report a mean pseudo-
thermodynamic property. Ultimately, the two values provide
high (dynamic) and low (pseudo-thermodynamic) bounds to
a linearly weighted mean film T g , with instrument-specific
weightings potentially playing a role in addition to these
generic effects.58 In our related paper,94 we additionally illus-
trate that, moving beyond nanoconfinement effects on single
systems, these metrology-dependent weighting effects can
confound observed trends in the strength of nanoconfinement
effects from system to system. Future work should therefore
employ pseudo-thermodynamic and dynamic methods as com-
plementary probes of nanoconfinement effects on segmental
dynamics and the glass transition.
SUPPLEMENTARY MATERIAL
See supplementary material for additional data regarding
the determination of local and mean T g in these systems.
ACKNOWLEDGMENTS
The authors thank Bryan Vogt for valuable discussions
and editorial advice.
This material is based upon work supported by the
National Science Foundation under Grant No. DMR1310433.
This work was supported by an allocation of computing time
from the Ohio Supercomputer Center.
1 J. Forrest, K. Dalnoki-Veress, and J. Dutcher, Phys. Rev. E 56, 5705 (1997).
2 J.A. Forrest and J. Mattsson, Phys. Rev. E 61, R53 (2000).
3 J.A. Forrest and K. Dalnoki-Veress, Adv. Colloid Interface Sci. 94, 167
(2001).
4 C. L. Jackson and G. B. McKenna, J. Non-Cryst. Solids 131133(Part 1),
221 (1991).
5 G. B. McKenna, Eur. Phys. J.: Spec. Top. 189, 285 (2010).
6 R. Richert, Annu. Rev. Phys. Chem. 62, 65 (2011).
7 M. Alcoutlabi and G. B. McKenna, J. Phys.: Condens. Matter 17, R461
(2005).
203316-8 Mangalara et al.
8 J. Baschnagel and F. Varnik, J. Phys.: Condens. Matter 17, R851 (2005).
9 C. B. Roth and J. R. Dutcher, J. Electroanal. Chem. 584, 13 (2005).
10 M. D. Ediger and J. A. Forrest, Macromolecules 47, 471 (2014).
11 D. S. Simmons, Macromol. Chem. Phys. 217, 137 (2016).
12 K. H. Nagamanasa, S. Gokhale, R. Ganapathy, and A. K. Sood, Proc. Natl.
Acad. Sci. U. S. A. 108, 11323 (2011).
13 K. Arabeche, L. Delbreilh, J.-M. Saiter, G. H. Michler, R. Adhikari, and
E. Baer, Polymer 55, 1546 (2014).
14 R. J. Lang, W. L. Merling, and D. S. Simmons, ACS Macro Lett. 3, 758
(2014).
15 F. W. Starr and J. F. Douglas, e-print arXiv:0906.5275 [cond-mat] (2009).
16 F. W. Starr and J. F. Douglas, Phys. Rev. Lett. 106, 115702 (2011).
17 B. A. P. Betancourt, J. F. Douglas, and F. W. Starr, Soft Matter 9, 241 (2012).
18 F. Chen, A. Clough, B. M. Reinhard, M. W. Grinstaff, N. Jiang, T. Koga,
and O. K. C. Tsui, Macromolecules 46, 4663 (2013).
19 H. C. Wong, A. Sanz, J. F. Douglas, and J. T. Cabral, J. Mol. Liq. 153, 79
(2010).
20 C. M. Stafford, B. D. Vogt, C. Harrison, D. Julthongpiput, and R. Huang,
Macromolecules 39, 5095 (2006).
21 J. M. Torres, C. M. Stafford, and B. D. Vogt, Acs Nano 3, 2677 (2009).
22 P. A. OConnell, S. A. Hutcheson, and G. B. McKenna, J. Polym. Sci., Part
B: Polym. Phys. 46, 1952 (2008).
23 X. Zheng, M. H. Rafailovich, J. Sokolov, Y. Strzhemechny, S. A. Schwarz,
B. B. Sauer, and M. Rubinstein, Phys. Rev. Lett. 79, 241 (1997).
24 D. Ruan and D. S. Simmons, Macromolecules 48, 2313 (2015).
25 D. Ruan and D. S. Simmons, J. Polym. Sci., Part B: Polym. Phys. 53, 1458
(2015).
26 C. B. Roth and J. M. Torkelson, Macromolecules 40, 3328 (2007).
27 M. Z. Slimani, A. J. Moreno, and J. Colmenero, Macromolecules 44, 6952
(2011).
28 M. Z. Slimani, A. J. Moreno, and J. Colmenero, Macromolecules 45, 8841
(2012).
29 C. J. Ellison and J. M. Torkelson, Nat. Mater. 2, 695 (2003).
30 R. D. Priestley, C. J. Ellison, L. J. Broadbelt, and J. M. Torkelson, Science
309, 456 (2005).
31 K. Paeng, R. Richert, and M. D. Ediger, Soft Matter 8, 819 (2011).
32 P. Z. Hanakata, J. F. Douglas, and F. W. Starr, J. Chem. Phys. 137, 244901
(2012).
33 R. J. Lang and D. S. Simmons, Macromolecules 46, 9818 (2013).
34 P. Scheidler, W. Kob, and K. Binder, Europhys. Lett. 59, 701 (2002).
35 P. Scheidler, W. Kob, and K. Binder, Eur. Phys. J. E 12, 5 (2003).
36 P. Scheidler, W. Kob, and K. Binder, J. Phys. Chem. B 108, 6673 (2004).
37 Y. Miwa, T. Kondo, and S. Kutsumizu, Macromolecules 46, 5232 (2013).
38 J. L. Keddie, R. A. L. Jones, and R. A. Cory, Europhys. Lett. 27, 59 (1994).
39 C. L. Jackson and G. B. McKenna, Chem. Mater. 8, 2128 (1996).
40 C. J. Ellison, M. K. Mundra, and J. M. Torkelson, Macromolecules 38, 1767
(2005).
41 C. J. Ellison, R. L. Ruszkowski, N. J. Fredin, and J. M. Torkelson, Phys.
Rev. Lett. 92, 095702 (2004).
42 R. A. Riggleman, K. Yoshimoto, J. F. Douglas, and J. J. de Pablo, Phys.
Rev. Lett. 97, 045502 (2006).
43 P. Z. Hanakata, J. F. Douglas, and F. W. Starr, Nat. Commun. 5, 4163 (2014).
44 J. E. Pye, K. A. Rohald, E. A. Baker, and C. B. Roth, Macromolecules 43,
8296 (2010).
45 A. Serghei and F. Kremer, Rev. Sci. Instrum. 77, 116108 (2006).
46 M. Tress, M. Erber, E. U. Mapesa, H. Huth, J. M uller, A. Serghei, C. Schick,
K.-J. Eichhorn, B. Voit, and F. Kremer, Macromolecules 43, 9937 (2010).
47 F. Kremer, M. Tress, and E. U. Mapesa, J. Non-Cryst. Solids 407, 277
(2015).
48 M. Erber, M. Tress, E. U. Mapesa, A. Serghei, K.-J. Eichhorn, B. Voit, and
F. Kremer, Macromolecules 43, 7729 (2010).
49 A. Serghei, H. Huth, C. Schick, and F. Kremer, Macromolecules 41, 3636
(2008).
50 A. Serghei and F. Kremer, Macromol. Chem. Phys. 209, 810 (2008).
J. Chem. Phys. 146, 203316 (2017)
51 S. Kim, M. K. Mundra, C. B. Roth, and J. M. Torkelson, Macromolecules
43, 5158 (2010).
52 Z. Fakhraai and J. A. Forrest, Phys. Rev. Lett. 95, 025701 (2005).
53 C. Ye, C. G. Weiner, M. Tyagi, D. Uhrig, S. V. Orski, C. L. Soles, B. D. Vogt,
and D. S. Simmons, Macromolecules 48, 801 (2015).
54 J. A. Forrest, J. Chem. Phys. 139, 084702 (2013).
55 J. A. Forrest and K. Dalnoki-Veress, ACS Macro Lett. 3, 310 (2014).
56 S. Kim, S. A. Hewlett, C. B. Roth, and J. M. Torkelson, Eur. Phys. J. E 30,
83 (2009).
57 J. E. G. Lipson and S. T. Milner, Eur. Phys. J. B 72, 133 (2009).
58 C. Rotella, M. W ubbenhorst, and S. Napolitano, Soft Matter 7, 5260 (2011).
59 S. Mirigian and K. S. Schweizer, J. Chem. Phys. 141, 161103 (2014).
60 S. Mirigian and K. S. Schweizer, J. Chem. Phys. 143, 244705 (2015).
61 J. E. G. Lipson and S. T. Milner, Macromolecules 43, 9874 (2010).
62 K. Kremer and G. S. Grest, J. Chem. Phys. 92, 5057 (1990).
63 F. Varnik, J. Baschnagel, and K. Binder, Phys. Rev. E 65, 021507 (2002).
64 J. Buchholz, W. Paul, F. Varnik, and K. Binder, J. Chem. Phys. 117, 7364
(2002).
65 J.-L. Barrat, J. Baschnagel, and A. Lyulin, Soft Matter 6, 3430 (2010).
66 C. M. Evans, H. Deng, W. F. Jager, and J. M. Torkelson, Macromolecules
46, 6091 (2013).
67 M. E. Mackura and D. S. Simmons, J. Polym. Sci., Part B: Polym. Phys. 52,
134 (2014).
68 L. M. Hall, M. J. Stevens, and A. L. Frischknecht, Phys. Rev. Lett. 106,
127801 (2011).
69 G. S. Grest and K. Kremer, Phys. Rev. A 33, 3628 (1986).
70 R. A. Riggleman, J. F. Douglas, and J. J. de Pablo, Soft Matter 6, 292 (2010).
71 R. A. Riggleman, J. F. Douglas, and J. J. de Pablo, J. Chem. Phys. 126,
234903 (2007).
72 F. W. Starr, J. F. Douglas, and S. Sastry, J. Chem. Phys. 138, 12A541 (2013).
73 D. S. Simmons and J. F. Douglas, Soft Matter 7, 11010 (2011).
74 D. S. Simmons, M. T. Cicerone, Q. Zhong, M. Tyagi, and J. F. Douglas,
Soft Matter 8, 11455 (2012).
75 J. A. Torres, P. F. Nealey, and J. J. de Pablo, Phys. Rev. Lett. 85, 3221 (2000).
76 M. D. Marvin, R. J. Lang, and D. S. Simmons, Soft Matter 10, 3166
(2014).
77 E. C. Glor and Z. Fakhraai, J. Chem. Phys. 141, 194505 (2014).
78 E. C. Glor, R. J. Composto, and Z. Fakhraai, Macromolecules 48, 6682
(2015).
79 F. W. Starr, T. B. Schroder, and S. C. Glotzer, Macromolecules 35, 4481
(2002).
80 R. S. Hoy, K. Foteinopoulou, and M. Kr oger, Phys. Rev. E 80, 031803
(2009).
81 J.-X. Hou, C. Svaneborg, R. Everaers, and G. S. Grest, Phys. Rev. Lett. 105,
068301 (2010).
82 S. Plimpton, J. Comput. Phys. 117, 1 (1995).
83 M. E. Tuckerman, G. J. Martyna, and B. J. Berne, J. Chem. Phys. 93, 1287
(1990).
84 L. Martnez, R. Andrade, E. G. Birgin, and J. M. Martnez, J. Comput.
Chem. 30, 2157 (2009).
85 J.-P. Hansen and I.R. McDonald, Theory of Simple Liquids (Academic Press,
2006).
86 W. L. Merling, J. B. Mileski, J. F. Douglas, and D. S. Simmons, Macro-
molecules 49, 7597 (2016).
87 F. Kohlrausch, Pogg. Ann. Phys. 119, 352 (1863).
88 G. Williams and D. C. Watts, Trans. Faraday Soc. 66, 80 (1970).
89 H. Vogel, Phys. Zeit. 22, 645 (1921).
90 G. S. Fulcher, J. Am. Ceram. Soc. 8, 339 (1925).
91 G. Tammann and W. Hesse, Z. Anorg. Allg. Chem. 156, 245 (1926).
92 J. H. Mangalara and D. S. Simmons, ACS Macro Lett. 4, 1134 (2015).
93 S. Peter, H. Meyer, J. Baschnagel, and R. Seemann, J. Phys.: Condens.
Matter 19, 205119 (2007).
94 J. H. Mangalara, M. D. Marvin, N. R. Wiener, M. E. Mackura, and D.
S. Simmons, J. Chem. Phys. 146, 104902 (2017).